An AI agent can finish a task cleanly and still be wrong: bad tool choice, weak retrieval, an inflated cost path, with no error metric ever flagging it. Amazon's new CloudWatch Omni traces agent reasoning end to end, evaluates correctness and tool selection, and plugs into LangChain, CrewAI, Bedrock AgentCore and rival evaluators like Braintrust and Ragas.

— via InfoQ, Sergio De Simone